[llvm] [AMDGPU] Add i1 mul patterns (PR #67291)

[Pierre van Houtryve via llvm-commits]

Pierre-vh wrote:

> Would be better off legalizing these rather than selecting

We already have identical patterns for other operations like add, what makes `mul` different in this case?
If we legalize, do you mean just legalize to i32 or do we need some other lowering logic that's aware of divergence?

https://github.com/llvm/llvm-project/pull/67291